Inside a Fortune 50 healthcare payor, two thousand analysts used to spend their mornings on the same problem - reading documents, chasing context, opening the same dozen tabs. Today, they ask a question and an AI insight engine answers in minutes. The engine was built by Distyl AI. Most of the people who use it have never heard the name.
That is, more or less, the whole pitch. Distyl does not sell wonder. It sells the boring middle - the part where a frontier model meets a procurement department and somehow survives the introduction. It is not a chatbot company. It is not a foundation model company. It is the company that shows up in the conference room with engineers, eats the legacy schema, and leaves a working system behind.
The address, for the curious, is 55 Hawthorne Street in San Francisco, a few blocks from the bay and a long way from the demo culture of the rest of the industry. Inside, a roughly two-hundred-person team divides its time between the office and the floors of customer buildings - hospital systems in the midwest, factory operations in the south, telco headquarters spread across three time zones. The Slack channels read like a deployment log, not a product roadmap. That is on purpose.
01 / The problem they sawThe 95 percent that never shipped
Most enterprise AI dies somewhere between the steering committee and the staging environment. Industry trackers put the failure rate of corporate AI pilots at roughly nineteen out of twenty. Money gets spent. Slides get made. Production never happens. The pilots that do ship arrive late, brittle, and politely ignored by the people they were meant to help.
Arjun Prakash and Derek Ho watched this play out from inside Palantir, where forward-deployed engineers had spent a decade learning the unsexy art of installing software inside other people's organizations. They left in 2022 with a thesis that sounds almost dull written down: the model is not the bottleneck. The operating model is.
The diagnosis is not novel. McKinsey has been writing variations of it in glossy quarterlies for two years. What is novel is the prescription. Most consultancies hand you a roadmap and walk out. Most startups hand you an API and call it a partnership. Distyl picked an awkward third option - stay, build, run, get paid when it works.
02 / The founders' betForward-deployed, again
The bet was straightforward and unfashionable. Hire fewer prompt engineers, more operators. Send them inside the customer. Make them sit through the meetings. Pay them when the work works.
The first cheque arrived in April 2023 - a $7M seed from Khosla Ventures and Lightspeed, paired with a Services Alliance with OpenAI. The seed was small. The alliance was the asset. It meant that when a Fortune 100 board asked the obvious question - "who do we call?" - OpenAI had an answer with a phone number on it.
03 / The productDistillery, and a small army
The platform is called Distillery. Its job, as the name implies, is to take the messy fluid of an enterprise - documents, systems, tribal knowledge, the spreadsheet that runs supply chain - and reduce it to a context layer a frontier model can actually reason over. Around Distillery sits a second product: people. Forward-deployed teams who spend months inside a customer redesigning the work itself before they install a model anywhere near it.
There is also a diagnostic offering called Distyl Scans, which produces what the company describes as an "AI Readiness" score and a roadmap pointing at the highest-ROI deployments first. It is, with a straight face, a consultancy product about whether you should hire a consultancy. Wilde would have approved.
The team itself is the third product, whether the marketing site admits it or not. Engineers and operators pulled from Palantir, Apple, BlackRock, Citadel, Snorkel, and a small number of frontier labs. People who have shipped software inside large institutions and lived to tell about it. People comfortable with the meeting where legal asks where the data went, and the meeting where finance asks why the bill is shaped like this.
Milestones, mostly recent
04 / The proofNumbers, on the record
A company can claim a lot in a press release. Distyl's claims are unusually specific. A Fortune 50 healthcare payor saved more than $23 million annually after the team rebuilt its investigations workflow around an AI insight engine. A hardware manufacturer cut supply-chain root-cause analysis time by roughly 80 percent. A separate Fortune 500 operator estimated a 47 percent improvement in daily task resolution time. Revenue, by the company's own account, grew 5x in 2024 and is tracking 8x in 2025.
Three deployments. Three numbers Distyl will say in public.
The cap table, for what it is worth, treats those receipts as real. Khosla Ventures and Lightspeed have written checks at every stage. Coatue, Dell Technologies Capital and DST Global joined at Series B. Valuation jumped roughly nine-fold between the November 2024 round and September 2025. That is fast, even by the standards of a year when "fast" is doing a lot of work.
05 / The missionAbundant intelligence, human agency
The official line is that Distyl helps the world's institutions rearchitect their systems for abundant intelligence from AI - without sacrificing human agency and control. Stripped of the prose, it means two things at once. First, that frontier models are about to make a lot of decisions, and somebody had better be in the room when those decisions get wired into a hospital, a factory floor, a customer service desk. Second, that the consultancy model itself - the slide deck, the recommendation, the polite exit - is the wrong shape for this moment. You cannot recommend an AI strategy. You have to build it, run it, and stay long enough to be blamed if it breaks.
Which is why Distyl prices on outcomes where it can. Why it does not write code in PowerPoint. Why its engineers move into the customer's tools, not the other way around. And why, in conversations with reporters, the founders keep using the word "production" the way other founders use the word "vision."
06 / Why it matters tomorrowThe decade of the deployment
Foundation models are not the scarce resource anymore. The scarce resource is the ability to plug them into a regulated industry without losing a finger. Healthcare payors, manufacturers, banks, telcos - the institutions that move the actual economy - cannot rip and replace. They have to retrofit. Quietly. In production. With auditability the legal team can defend.
That is the market Distyl is betting on, and the market that, for once, seems to be betting back. The next twelve months will test whether a forward-deployed model can scale past its founders, whether outcome-based pricing survives contact with procurement, and whether the company's 100 percent deployment record holds when the deployments number in the hundreds instead of the dozens.
There are reasonable doubts. Services businesses are famously hard to scale without diluting quality, and the consulting graveyard is full of firms that confused the first ten engagements with a repeatable system. Outcome-based pricing has a way of becoming a creative-accounting exercise once general counsel notices it. And the OpenAI alliance, while a powerful door-opener, is shared - Distyl is one name on a list, not the list. None of that is fatal. All of it is real.
The argument for Distyl is that the alternative is worse. The Fortune 500 cannot wait for the perfect platform; the platform companies cannot, in any reasonable time frame, install themselves inside three hundred regulated enterprises. Somebody has to do the carpentry. Distyl has volunteered, hired the carpenters, and started shipping the cabinets.
07 / Back to the morningThe analyst, again
Return, for a moment, to that healthcare payor at the top of the page. The analyst still arrives at the same desk. The same monitor. The same coffee, possibly worse. She asks her question. The answer is on her screen before she has finished typing it. She has no idea who Distyl is, and there is something almost dignified about that. The point was never the logo. The point was the morning.